COMPARING THREE CANOPY REFLECTANCE MODELS WITH HYPERSPECTRAL MULTIANGULAR SATELLITE DATA - PDF document

COMPARING THREE CANOPY REFLECTANCE MODELS WITH HYPERSPECTRAL MULTI-ANGULAR SATELLITE DATA M. Schlerf a, *, W. Verhoef a, b, H. Buddenbaum c, J. Hill c, C. Atzberger d, A. Skidmore aa International Institute for Geo-Information Science and Earth Observation, P.O. Box 6, 7500 AA Enschede, The Netherlands b National Aerospace Laboratory NLR, P.O. Box 153, 8300 AD Emmeloord, The Netherlands c Remote Sensing Department, University of Trier, Behringstrasse, D-54286 Trier, Germany d Corresponding author: schlerf@itc.nl 1.INTRODUCTION Among various methods, physically based radiative transfer models have proved to be a promising alternative to estimate biophysical vegetation attributes. However, the existence of many canopy reflectance models with different levels of complexity makes it difficult to choose the most appropriate was 50° (Figure 1). All five images were geocoded to the local reference system using image-to-image registration and resampled to the nominal ground resolution at nadir view (34 m). The images were radiometrically corrected to top-of-canopy reflectance assuming standard atmospheric parameters. Figure 1: Observation and sun geometry of Chris/PROBA images acquired over Idarwald test site on 05/09 2005 Figure 2: Bi-directional Chris/PROBA spectral reflectances of spruce stands after image radiometric correction. Each curve corresponds to the average signature of 15 stands (60 pixels in total). Beech stands exhibit similar directionality. Two weeks after the image acquisition, an extensive field campaign was conducted and the canopy structure of altogether 28 forest stands (15 plots of Norway spruce (Picea abies L. Karst.)) and 13 plots of Beech (Fagus sylvatica)) was measured. In each forest stand, a plot of 30x30 m size was established and its central position determined using a hand-held GPS device. Measured attributes include leaf area index (LAI, measured with Li-Cor LAI-2000), crown diameter (CD), tree density (SD), crown cover (COcr), canopy height (TH), and percent coverage of understorey vegetation (COus). 2.3Model parameterisation We identified three typical stands for each species (Table 1): young, medium, and old stands. Young stands (20-30 years; before thinning) typically have small tree crowns, many trees and relatively large canopy LAI, whereas old stands (� 80 years) are usually thinned out and therefore show lower LAI values and larger crowns. At first, for all of the six stands, an average hyperspectral signature was extracted from the four nearest pixels surrounding the GPS-measured plot location in each Chris/PROBA image The angular variations of the spectral signatures of spruce stands is illustrated in Figure 2. Type Age ID Spruce 21 83 Spruce 42 109 Spruce 131 301 Beech 40 8 Beech 84 243 Beech 122 245 Table 1: Selected forest stands. Age = stand age in 2005 (years), ID = forest stand identification ID CD TH SD LAIc LAIs CO cr COus 83 1.75 7.9 5000 5.1 7.25 80 30 109 4.30 15.2 689 3.8 6.00 70 70 301 6.60 35.4 211 3.9 7.71 35 40 8 3.03 13.2 3022 6.1 6.59 85 10 243 5.70 24.8 467 3.4 4.80 50 10 245 7.85 25.2 233 4.3 6.29 35 20 Table 2: In situ measured canopy properties. CD = crown diameter (m), TH = tree height (m), SD = stem or tree density (/ha), LAIc = canopy leaf area index (Li-Cor LAI-2000), LAIs = single tree leaf area index (LAIc/COcr), COcr = crown cover (%, visually estimated), COus = cover of understorey vegetation (%, visually estimated) Spruce Beech N 3.0 1.5 Cab 70 50 Cw 0.02 0.02 Cm 0.025 0.015 ala 65 50 hot 0.02 0.04 Table 3: Leaf properties and species specific canopy properties. N = leaf structure parameter, Cab = chlorophyll a+b content (µg/cm²), Cw = equivalent water thickness (g/cm²), Cm = dry matter content (g/cm), ala = average leaf angle (º), hot = hot spot parameter Figure 3: INFORM simulations of a spruce canopy in nadir direction (stand 301). R_soil (soil and litter) = scaled HyMap spectrum, R_understorey = R_soil + understorey vegetation (LAI=0.5), R_infinite = Tree crown LAI = 15, R_forest = Forest canopy reflectance To allow comparison of modelled and measured reflectance, the models were parameterised to a large extent using the ground truth information that had been measured in the respective forest stands. Ground truth was available for the most important structural canopy parameters (Table 2). Leaf properties and certain canopy properties that have not been measured in the field (Table 3) were fixed to species specific default values or were slightly optimised through comparing INFORM model outputs with Chris reflectances. All models used a similar background spectrum (Figure 3, black dashed line named understorey). Specification of the sun and viewing geometry was matched to the image acquisition (section 2.2). Using the stand type specific canopy attributes and the illumination and viewing geometries of the Chris/PROBA overpass, the canopy reflectance was modelled using the three radiative transfer models. In FRT, additional input parameters had to be defined. The following input parameters were computed from measured values: crown length (m), trunk diameter (m), total dry leaf weight (DLW, kg/tree), chlorophyll-% of DLW, water-% of DLW, dry matter-% of DLW, among others. For the following input parameters default values were used: shoot shading coefficient, refraction index ratio, shoot length, Markov parameter, among others. Cones (spruce) and ellipses (beech) were assigned to approximate crown shape. 3.RESULTS Comparison of the directional reflectances simulated by FRT, SLC, and INFORM reveals general agreement but also some differences among the models. In general, the models produce signatures comparable to measured ones for spruce and beech forests of different age and canopy structure. The pronounced bowl shape in the red waveband of the Chris/PROBA data (Figure 4) is not followed by the models; they rather produce a straight line. Possibly, in the atmospheric correction of the satellite data, the value for the aerosol optical thickness should have been higher, yielding lower reflectances in the red, and thus less of a bowl shape. Figure 4: Measured and simulated bi-directional reflectances in the red (670 nm) for spruce forests (left) and beech forests (right) labelled with the respective stand ID. Grey line = FRT, black dashed line = INFORM, black dotted line = SLC, black stars = Chris/PROBA. Figure 5: Directional reflectances in the NIR (780 nm) for spruce forest (left) and beech forest (right) labelled with the respective stand ID. Grey line = FRT, black dashed line = INFORM, black dotted line = SLC, black stars = Chris/PROBA. View zenith angle (º) View zenith angle (º) FRT underestimates red reflectances of young forest stands at large negative viewing angles whereas model outputs of SLC and INFORM are pretty similar to the measured ones. In the NIR (Fig. 5), differences are more pronounced than in the red. SLC overestimates NIR reflectance for older stands, in particular spruce but also beech. Approximating the Chris/PROBA reflectance curves with SLC for those stands would be possible when a darker background is used or woody material is included. Similar to the red domain, FRT underestimates NIR reflectances at large negative viewing angles. FRT and SLC in general show a relatively low directional variablility in the NIR compared to the INFORM model whose outputs better approximate the measured reflectances. In this study, parameters that were not measured (e.g. ala, Cm) were optimised using INFORM; SLC and FRT had to follow INFORM in those parameter which partly explains why INFORM shows the best results. In the NIR, FRT performs slightly better than SLC as it has additional free parameters that were modified to approximate the satellite measuredreflectances. 4.CONCLUSIIONS This is one of the first studies that compare modelled reflectances with satellite-measured reflectances of forest canopies for different wavelengths and view angles. The results are promising, as there is a general agreement between modeled and measured spectra. Most of the differences among the models can be explained by the inherent model structure or by the way the models were parameterized. The results of this research represent an important step towards model development, understanding of radiative transfer in canopies and subsequent model inversion. 5.ACKNOWLEDGEMENTS The authors would like to thank Bianca Hoersch (European Space Agency, ESA) and Peter Fletcher (Remote Sensing Applications Consultants, RSAC) for acquiring Chris/PROBA data over Idarwald Forest and Andres Kuusk (Tartu Observatory) for providing the FRT model. 6.REFERENCES ATZBERGER, C. (2000): Development of an invertible forest reflectance model: The INFOR model.In: Buchroithner, M. (Ed.): Proceedings of the 20th EARSeL Symposium Dresden, Germany, 14-16 June 2000: 39-44. BACOUR, C., JACQUEMOUD, S., TOURBIER, Y., DECHAMBRE, M., FRANGI, J.-P. (2002): Design and analysis of numerical experiments to compare four canopy reflectance models. Remote Sensing of Environment, 79: 72-83. BARET, F., FOURTY, T. (1997), Estimation of leaf water content and specific leaf weight from reflectance and transmittance measurements. Agronomie 17:455–464. JACQUEMOUD, S., BACOUR, C., POILVE, H., FRANGI, J.-P. (2000): Comparison of four radiative transfer models to simulate plant canopies reflectance: Direct and inverse mode. RemoteSensing of Environment, 74: 471-481. KUUSK, A., (2001). A two-layer canopy reflectance model. Journal of Quantitative Spectroscopy & Radiative Transfer 71: 1–9. KUUSK, A., NILSON, T. (2000): A directional multispectral forest reflectance model. Remote Sensing of Environment 72: 244-252. NILSON, T., PETERSON, U., (1991). A Forest Reflectance Model and a Test Case. Remote Sens. Environ. 37: 131–142. PINTY B, WIDLOWSKI JL, TABERNER M, ET AL. (2004): Radiation Transfer Model Intercomparison (RAMI) exercise: Results from the second phase. Journal of Geophysical Research - Atmospheres 109 (D6). PRICE, J.C., (1990). On the information content of soil reflectance spectra. Remote Sens. Environ. 33: 113–121. ROSEMA, A., VERHOEF, W., NOORBERGEN, H., & BORGESIUS, J. J. (1992). A new forest light interaction model in support of forest monitoring. Remote Sensing of Environment, 42, 23– 41. SCHLERF, M. & ATZBERGER, C. (2006): Inversion of a forest reflectance model to estimate structural vegetation attributes using hyperspectral remote sensing data. Remote Sensing of Environment, 100: 281-294. VERHOEF, W. (1984). Light scattering by leaf layers with application to canopy reflectance modeling: The SAIL model. Remote Sensing of Environment, 16, 125– 141. VERHOEF, W. (1985). Earth observation modeling based on layer scattering matrices. Remote Sensing of Environment, 17, 165– 178. VERHOEF, W. (2002): Improved modeling of multiple scattering in leaf canopies: the model SAIL++. In: Sobrino, J.A. (Ed.), Proc. 1st Int. Symp. on Recent Advances in Quantitative Remote Sensing, Universitat de Valencia, Spain: 11-20. VERHOEF, W., BACH, H. (2003): Simulation of hyperspectral and directional radiance images using coupled biophysical and atmospheric radiative transfer models. Remote Sensing of Environment, 87: 23/41. VERHOEF, W. & BACH, H. (2007): Coupled soil-leaf-canopy and atmosphere radiative transfer modeling to simulate hyperspectral multi-angular surface reflectance and TOA radiance data, Remote Sensing of Environment, in press.